Anthropic has abruptly disabled access to its most powerful artificial intelligence models, Claude Fable 5 and Claude Mythos 5, following a direct order from the United States government. The company confirmed that it received an export control directive on Friday, June 12, 2026, at 5:21 p.m. ET, which mandates a complete suspension of access to these specific models for all foreign nationals, whether they are located inside or outside the country.
Because the government’s directive applies to such a wide range of individuals, including foreign national employees working directly for Anthropic, the company decided it was impossible to maintain selective access. To ensure full compliance with the law, Anthropic took the precautionary step of disabling these models for all customers globally. The company clarified that this unexpected shutdown does not affect access to any of its other AI models.
The sudden move follows the release of Claude Fable 5 just days earlier, on June 9. While the government letter did not provide comprehensive details regarding its national security concerns, Anthropic stated its understanding that the order stems from a specific method of bypassing, or “jailbreaking,” the Fable 5 safety controls. According to the company, this method involves asking the model to read a specific codebase to identify and fix software flaws.
Anthropic maintains that this potential vulnerability is minor and already well-documented. In its internal review, the company found that the same level of capability is widely available through other publicly deployed models, such as OpenAI’s GPT-5.5. Anthropic officials argued that using this standard to pull a commercial model from hundreds of millions of users would effectively halt all new model deployments across the entire frontier AI industry.
Despite these objections, Anthropic is complying with the directive while working to resolve what it describes as a misunderstanding. The company expressed its disagreement with the government’s assessment, noting that the reported jailbreak method is a routine task performed by cybersecurity defenders every day to keep critical systems secure. Anthropic representatives emphasized that they are committed to restoring access as quickly as possible.
Before this abrupt suspension, Fable 5 and Mythos 5 were positioned as the most capable AI models in the company’s history. Fable 5 was designed as a “Mythos-class” model for general use, featuring a 1 million token context window and high-level reasoning capabilities for complex, long-running tasks. Mythos 5, its restricted counterpart, remained available only to vetted partners through Project Glasswing, an initiative focused on helping cyber defenders secure essential infrastructure.
The impact of this shutdown is significant, as thousands of developers and researchers had begun integrating Fable 5 into ambitious projects ranging from large-scale code migrations to advanced scientific research. As the industry watches for updates, questions remain about how the U.S. government will navigate the fine line between preventing potential security risks and stifling the rapid advancement of frontier AI technology.
